Application of genetic stochastic resonance algorithm to quantitative structure–activity relationship study

Abstract Based on the theory of stochastic resonance (SR), a genetic stochastic resonance (GSR) algorithm is proposed and employed in a quantitative structure–activity relationship (QSAR) study. In the GSR algorithm, the variables that are related to the bioactivity of a molecule series are considered as ‘signal’ and the other nonrelated features as ‘noise’. The ‘signal’ is amplified in a nonlinear system with the optimized parameters by SR. The optimization of the parameters is supervised by a specified bioactivity in GSR using genetic algorithm (GA). The GSR algorithm was investigated with a published data set. The relevant variables are enhanced and their power spectra are significantly changed and similar to that of the bioactivity after GSR. The descriptor matrix consequently becomes more informative and the collinearity is suppressed. Therefore, the coming procedure of feature selection becomes easier and more efficient. The linear QSAR models of the data set obtained by GSR have better performances and are more predictive than those by the reported approaches. It is demonstrated that GSR is an effective tool in QSAR study.